Keynote: Data and Programs: Rethinking the Fundamentals
This was a pretty big disaster. Don Box had a bad day. As far as I’m concerned, that’s all it was.
The feedback system was GREEN, YELLOW and RED cards that you deposit on the way out. There were lots of reds. I declined participation; he knew it bombed. He didn’t need us to tell him.
“Patterns for Cloud Computing” – Simon Guest
This was excellent. It made me want to run out and start using Windows Azure.
Simon Guest is a Microsoft guy, but I found the presentation to be well balanced regardless. Of course, he showed off Azure. But, he also spoke quite a bit about Google and about Amazon. He didn’t bash them or even compare them positively or negatively to Azure. He just talked about them and what they do.
In one demo, he submitted a request to the cloud. The page came back without the row because it hadn’t committed yet. A second later, the grid regenerated and the row was there. This brought me back to the WEBSOCKETS talk… whatever polling or refreshing that was doing can be eliminated when the websocket is implemented. (See Wednesday – HTML 5)
This covered the basics of Azure, which I already knew from some reading: submit a request; put the request on a queue; process the queue with one or more workers. (This is how a product we built at work does things too).
The thing that surprised me about it is that you control the scaling. You tell it how many workers to use. If you get more load, you can increase the number of workers. I figured you’d be able to apply a policy or something to control this so that it can automatically scale up as it needs to within boundaries that you define. (Maybe it does do that, but that’s not what we saw).
Another demo was “find all of the prime numbers between 1 and x”. He went all out with that one… he distributed the calculations across Azure, Google and Amazon. The neatest part is that it reported it took a total of 6 CPU seconds across all of the machines. Neat.
He also touched on SQL SERVICES for AZURE, which weren’t originally there. Its not very cloudish, but there was a lot of demand for it, so they provided it. You can setup a 1 gig database or a 10 gig database. All databases are replicated across at least 3 servers.
Despite that its not very cloudish, it seems that will make Azure more practical for a lot more existing applications.
“Mapping Relational Data Model Patterns to the App Engine Data Store” – Max Ross, Google
Max was brutally honest about the capabilities of the Google app engine to the extent that he talked a lot about things that it can’t do. I ran through the app engine startup tutorials probably a year ago, but that was about it. I couldn’t relate to the discussion, though I can say it was lively. People in the room had good questions.
“Architecting for the Cloud: Horizontal Scaleability via Transient, Shardable, Share-Nothing Resources” – Adam Wiggins
Another great talk that I couldn’t relate to. HEROKU is a cloud platform for Ruby applications. I’m looking at the website now, and it has “40,653 apps running right now!”.
Unfortunately, I don’t have anything else to say about this one. I’m note a ruby guy, but if I were, HEROKU would be the place to go.
“Agile Development to Agile Operations” – Stuart Charlton
Disclaimer: I’m not sure if I’m remembering this accurately or confusing it with something else. Feel free to disregard. (I didn’t take notes, and my memory is fuzzy))
This talk was about the changes to infrastructure when using a cloud based system.
Without a cloud, you need enough servers to cover all of your peaks. With a cloud, you only need to scale up when necessary. Only use the servers you need when you need them.
Amazon’s cloud solution is good for this. He gave an example of when someone had to convert millions of TIFFs to some other format (PDF, perhaps?). They were able to do it in a weekend using Amazon’s services. Rather than having to obtain all of the hardware and set it up, they used Amazon, and got it all done in an absurdly small amount of time.